A Montreal company has developed software for the automatic detection of marine mammals from aerial, satellite and boat images. A new study validates his approach.
“Fisheries and Oceans gave us a big contract to review 100,000 aerial images in the Arctic,” says Emily Charry Tissier, CEO of Whale Seeker. “We look at all marine mammals, but our study focuses on belugas in 5334 images. »
Posted in Frontiers in Marine Science, the study concludes that the success rate of the AI algorithm is slightly higher than that of humans. Hit rate is defined as the agreement between detection of animals by AI and detection by human observers.
“We had the idea for the project in 2018, explains Mr.me Charlie Tissier. Bertrand Charry and I are both biologists. We had a contract with WWF to detect narwhals in aerial images. There were a lot of differences between the sensing that each of us was doing. We saw the need to standardize the detection of marine mammals from images. »
PHOTO PROVIDED BY WHALE SEEKER
Bertrand Charry, Antoine Gagné Turcotte and Emily Charry Tissier, of the Montreal company Whale Seeker
The third co-founder, Antoine Gagné Turcotte, was then a programmer in a large company. “I was looking for another job that would give more importance to values,” says Mr. Gagné Turcotte. Talking with Emily at the Père-Marquette dog park, we decided to start the business. »
The name is in English because the market for this technology is global.
Many companies are interested in our product. It can help with environmental impact studies before a mining or infrastructure project.
Emily Charry Tissier, CEO of Whale Seeker
Competition
Are there competitors in the field? “There are researchers who use AI for marine mammal detection, but they use open source software. It is unclear what is in the box of the software, especially if it is trained by experts. »
Training AI software is a crucial step. In the case of the study, Whale Seeker’s software was trained with 100 images where human experts pointed to belugas.
At the moment, the results of the AI software are validated by humans. The software also asks them for help for photos where it is difficult to identify marine mammals. “I think that within one or two years, we will be able to limit the involvement of humans to difficult cases and random checks,” says Mr. Gagné Turcotte.
Satellites
The images may come from satellites, ground stations or ships. “With satellites, there’s a resolution limitation, so it’s harder to tell one species from another,” says Ms.me Charlie Tissier. On board ships, we will use infrared detection of the breath of marine mammals. »
Why not use sonar? “We specialize in passive detection, says Mme Charlie Tissier. A sonar creates noise, we find that there is already a lot of it in the ocean. »
The use of drones could enable real-time surveillance in commercially important sea areas. “We could, for example, close and open fishing areas more dynamically, to limit prohibitions to a minimum,” says Ms.me Charlie Tissier.
Learn more
-
- 900
- Number of belugas in the estuary
of the St. Lawrence
Source: Fisheries and Oceans Canada